No Black Box Anymore: Demystifying Clinical Predictive Modeling with Temporal-Feature Cross Attention Mechanism.

OBJECTIVES

The aim of this study is to assess prospectively the effect of rituximab (RTX) on MRI features of wrist joint disease in patients affected by rheumatoid arthritis (RA).

METHODS

Ten patients (6F/4M, mean age 52.9±15.5 years) diagnosed with IgM rheumatoid factor, anti-CCP positive, RA according to the 1987 ACR criteria were treated with a single course of RTX (2 infusions of 1000 mg, 15 days apart). MRI of the dominant hand was performed with a 0.2T extremity-dedicated machine using pre and post contrast T1 weighted SE, turbo 3D, and STIR sequences at baseline, and after 4 and 24 weeks. MRI was analysed using the OMERACT-RAMRIS score and the dynamic contrast-enhanced (DCE-MRI) technique for wrist synovitis, which calculates the enhancement ratio as both rate of early enhancement (REE) and relative enhancement (RE). The corresponding ME and IRE parameters were calculated also through a computer-aided semi-automated method on the mean of three MRI slices and on a small ROI positioned in the area of maximum enhancement.

RESULTS

DAS significantly decreased during the study period (ANOVA for repeated measures, p=0.005). The RAMRIS score did not change along the study, whereas the dynamic MRI values RE, IRE and ME on the small ROI significantly decreased. RE, but not the RAMRIS synovitis score, significantly correlated with DAS at baseline, 1 and 6 months (p=0.005, 0.04, and 0.0007, respectively).

CONCLUSIONS

RTX confirmed good clinical efficacy, which was paralleled by a significant decrease in dynamic MRI results for wrist synovitis. On the contrary, the traditional RAMRIS measures did not change.